Liver fibrosis scores predict liver disease mortality in the United States population

MATERIALS AND METHODS

The NHANES III was conducted in the United States from 1988 through 1994 by the National Center for Health Statistics of the Centers for Disease Control and Prevention.(27) The survey consisted of cross-sectional interview, examination, and laboratory data collected from a complex multistage, stratified, clustered probability sample representative of the civilian, noninstitutionalized population with oversampling of non-Hispanic blacks, Mexican Americans, and persons aged 60 years or older. The survey was approved by the Centers for Disease Control and Prevention ethics review board, and all participants provided written informed consent to participate.

Data were collected at baseline on factors believed to be related to liver injury and used to calculate fibrosis scores and/or included as covariates in multivariate analyses:(28–30) age, sex, race-ethnicity, education, alcohol intake, cigarette smoking, caffeine intake from beverages, physical activity, diagnosed diabetes, body mass index [weight (kg) / height (m²)]; waist-to-hip circumference ratio, and systolic and diastolic blood pressure. A venous blood sample was collected and shipped weekly at −20°C to testing laboratories. Aspartate aminotransferase, alanine aminotransferase, gamma glutamyltransferase, albumin, platelet count, hepatitis B virus surface antigen, hepatitis C virus RNA, transferrin saturation, and concentrations of hemoglobin A_1C, total and high-density lipoprotein cholesterol, C-reactive protein, alkaline phosphatase, and total bilirubin were measured as previously described.(31) Diabetes was defined as a health care provider diagnosis or hemoglobin A_1C ≥6.5% and prediabetes as hemoglobin A_1C ≥5.7-<6.5%.

Four scores, APRI, FIB-4, NFS, and Forns were calculated as previously described (Table 1).(16, 19, 22, 24) The originally described cut-points for each score were used to categorize liver fibrosis probability as low, intermediate, or high (Table 1).(18, 19, 22, 32) For APRI, intermediate and high categories were combined because of low liver disease mortality numbers in the NHANES III study sample representative of the U.S. general population.

Of 23,258 sampled persons aged 20 years and older in NHANES III, 16,573 (71%) attended a study visit at a mobile examination center. Participants were excluded if they were positive for serum hepatitis B surface antigen or hepatitis C RNA (n=346). We additionally excluded persons who were missing data on viral hepatitis serology (n=1,084) or fibrosis score components (for APRI and FIB-4, n=279 and for NFS, n=379), were ineligible for mortality follow-up (n=12), or did not have a cause of death available (n=11), resulting in analysis samples of 14,841 for APRI and FIB-4 and 14,741 for NFS. One component of the Forns score, gamma glutamyltransferase, was added to the protocol after the beginning of the survey (n=3,531 missing), resulting in an analysis sample of 11,589 (Figure 1).

Participants were passively followed for mortality through December 31, 2011, by linking NHANES III with National Death Index records through a probabilistic match, a well-established and validated matching method.(33, 34) Mortality outcomes were based on death certificate underlying or contributing cause coded according to the International Classification of Diseases, Ninth Revision (ICD-9), for deaths occurring between 1988 and 1998 and the International Classification of Diseases, Tenth Revision (ICD-10), for deaths occurring between 1999 and 2011. Outcomes consisted of all-cause mortality and cause-specific mortality from cardiovascular disease (ICD-9 codes 410–414, 428, 429.2, 433–435, 437.0–437.1, 440, and 444; ICD-10 codes G45, I20-I25, I50, I63, I65-I66, I67.2, I67.8, I69.3, I70, and I74), neoplasms (ICD-9 codes 140–154, 155.1, and 156–239; ICD-10 codes C00-C21, C22.1, and C23-D48), diabetes (ICD-9 code 250; ICD-10 codes E10-E14), and liver disease, including primary liver cancer (ICD-9 codes 155.0, 155.2, and 570–573; ICD-10 codes C22.0, C22.2-C22.9, and K70-K76). Restricted NHANES III mortality data were used for this analysis through the National Center for Health Statics Research Data Center.(35)

RESULTS

Distributions of APRI, FIB-4, NFS, and the Forns score among NHANES III participants without viral hepatitis, are shown by fibrosis score category in Table 2. Using originally described cut-points, prevalence of an intermediate and high fibrosis score varied (Table 2). In general, persons with higher scores had more adverse liver-related characteristics (Tables 3a and 3b). They were also older, less educated, and more likely to be obese, diabetic, and hypertensive, with higher C-reactive protein and lower caffeine intake (Tables 3a and 3b). Relationships with sex, race-ethnicity, alcohol consumption, cigarette smoking, physical activity, lipid levels, and transferrin saturation varied among fibrosis scores.

DISCUSSION

In this U.S. population-based study with up to 23 years of follow-up, intermediate to high liver fibrosis scores among persons without viral hepatitis were associated with increased liver disease mortality. This relationship was consistent in age- and multivariate-adjusted analyses and across four fibrosis scores, both non-specific and specific for NAFLD. Intermediate to high fibrosis scores were also associated with increased overall mortality in age- and multivariate-adjusted analyses. Associations with higher mortality from cardiovascular disease, neoplasms, and diabetes were found inconsistently across fibrosis scores.

Fibrosis scores have been shown to predict histological fibrosis with reasonable accuracy in cross-sectional studies.(16–24) Fibrosis scores may also be useful as prognostic tools among persons with NAFLD, but this has been less studied. Hepatic steatosis on ultrasound and elevated liver enzyme activities, as noninvasive markers of fatty liver or non-alcoholic steatohepatitis, have been examined in relation to disease outcomes, including mortality.(38, 39) However, there have been few prospective studies of fibrosis scores well powered to predict liver-disease mortality and other long-term adverse hepatic outcomes.(40–42) A previous analysis of the NHANES III population with mortality follow-up through 2006, found higher overall and cardiovascular disease mortality in association with three of the same fibrosis scores among persons with suspected NAFLD.(26) Higher liver disease mortality was not reported in that analysis which had shorter follow-up and fewer liver-related deaths. In an international, multicenter study of 320 biopsy-diagnosed NAFLD patients followed for a median of 8.7 years, fibrosis scores, including NFS, APRI, and FIB-4, identified patients at increased risk for liver-related complications, all-cause mortality, and liver transplantation.(40) In a community-based study of 302 NAFLD patients identified through the Rochester Epidemiology Project and followed for a mean of 12 years, a higher NFS at baseline was predictive of increased overall mortality risk.(41) The latter two studies demonstrated that fibrosis scores predicted all-cause mortality, but neither was large enough to investigate disease-specific mortality. We are unaware of other well-powered prospective studies of prognostic ability of fibrosis scores in NAFLD patient cohorts or the general population.

A diagnosis of ‘NAFLD’ requires exclusion of significant alcohol intake (>21 drinks on average per week for men or >14 drinks on average per week for women).(43) However, because over two thirds of U.S. adults are now overweight or obese, mixed forms of NAFLD and alcoholic fatty liver disease are increasingly likely.(44, 45) Furthermore, the accuracy of self-reported alcohol consumption is unknown and clinical diagnosis based on patient-reported intake remains a potential source of misclassification between modest and significant alcohol use. In this analysis, we did not exclude persons based on their reported alcohol consumption; instead, we adjusted for alcohol intake in multivariate analyses to present a more comprehensive picture of the fatty liver disease epidemic. Furthermore, because characteristic features of non-alcoholic steatohepatitis, such as steatosis and inflammation, may diminish as fibrosis progresses, we did not limit our analysis to persons with steatosis on ultrasound and/or elevated liver enzymes, but studied the entire NHANES III population who were negative for viral hepatitis markers.(46, 47)

A limitation of using NHANES III to study fibrosis scores is reliance on single measurements of serum markers without a histological diagnosis. However, liver biopsies cannot be conducted on the general population. Imaging methods that use transient elastography to measure liver stiffness and serum markers that directly measure components of extracellular matrix may be more accurate than indirect serum markers, but were not performed in NHANES III. In addition, the scoring systems we studied provide a qualitative, not quantitative measure of fibrosis, and have not been validated in the general population. However, these fibrosis scores have been validated in patients with viral hepatitis, alcoholic liver disease, NAFLD, and other diseases.

Furthermore, a recent report suggested that the performance of the NFS and FIB-4 for identifying and excluding advanced fibrosis may vary across the life span with lower accuracy among young adults and lower specificity among older adults.(48) Another limitation of our study was lack of validation of cause of death. Although ascertainment of vital status using the National Death Index is high (>99%), assigning cause of death based on death certificate diagnoses can lead to misclassification. Finally, because participants were not reevaluated during follow-up, some may have developed increased fibrosis, leading to misclassification. However, we hypothesize that an increase in fibrosis scores would lead to a stronger association with liver disease mortality. Likewise, we did not have any longitudinal data on changes in other baseline participant characteristics which could lead to misclassification during the follow-up period. Assuming this misclassification was non-differential among study participants, our results may be skewed toward null and underestimate the strength of the associations of liver fibrosis scores with mortality outcomes. These limitations are balanced by the benefits of a representative population-based sample, particularly avoidance of ascertainment bias found in clinical studies of conveniently selected patients, and ability to generalize results to the U.S. population. Other important strengths were over two decades of mortality follow-up and availability of data on numerous predictors of liver disease mortality measured at baseline.

In conclusion, in the U.S. population, intermediate to high liver fibrosis scores were associated with increased liver disease mortality. Higher overall mortality was also associated with elevated fibrosis scores; however, relationships were inconsistent with mortality from cardiovascular disease, neoplasms, and diabetes. A 2012 American Gastroenterological Association practice guideline on the diagnosis and management of NAFLD recommended the clinical use of the NFS for identifying NAFLD patients with higher likelihood of having bridging fibrosis and/or cirrhosis.(43) Our study provides support for the use of fibrosis scores, including the NFS, for identifying persons at high risk for advanced fibrosis.(25, 49) Therefore, liver health management with fibrosis risk stratification merits further investigation with serial monitoring and targeted interventions.